KaggleBench Dataset Ground Truth

I am looking through the paper and the dataset and trying to understand the ground truth. Are the answers in questions.json the ground truth or the agentada outputs. If they are the agentada outputs, where can I find the ground truth?